Qualcomm, a company primarily known for its dominance in mobile phone processors, has made a significant move into the lucrative artificial intelligence (AI) data center market. On Monday, October 27, 2025, the company unveiled a new series of chips designed to compete with industry leaders Nvidia and AMD for AI-related tasks, causing its stock to surge by over 20%.
This strategic pivot marks Qualcomm’s return to the data center space, aiming to capture a share of the unprecedented demand for powerful and efficient AI hardware. The company is leveraging its expertise in mobile chip technology to challenge the current market dynamics.
Introducing the AI200 and AI250
At the core of the announcement are two new AI accelerator solutions, the Qualcomm AI200 and Qualcomm AI250, which will be offered as accelerator cards and full, integrated rack-level systems.
The AI200 is scheduled for commercial release in 2026.
The AI250 is planned for the following year, 2027, and is expected to feature a groundbreaking new memory architecture.
These systems are designed to plug directly into data centers, providing rack-scale performance optimized for generative AI inference.
Technological Innovations and Key Features
Qualcomm’s new offerings are built on the company’s existing Hexagon Neural Processing Units (NPUs), which have been scaled up from its mobile processors to meet the demands of large-scale data centers. The company is focusing on energy efficiency and a lower total cost of ownership (TCO) as major selling points.
Key technical specifications include:
High Capacity Memory: The AI200 card supports a massive 768GB of LPDDR memory, providing higher capacity at a lower cost for AI inference tasks.
Advanced Memory Architecture: The upcoming AI250 will introduce an innovative architecture based on near-memory computing. Qualcomm promises this will increase effective memory bandwidth by more than 10 times while significantly reducing power consumption.
Rack Scale Systems: Both solutions will be available in integrated, direct liquid-cooled racks with a power consumption of 160 kW. They will feature PCIe for scaling up and Ethernet for scaling out.
Focus on AI Inference: The chips are specifically optimized for AI inference, the process of running finished AI models rather than the computationally intensive training phase.
Shifting the Competitive Landscape
Qualcomm’s entry introduces a new, formidable competitor into a market largely dominated by Nvidia, with AMD as a strong second. By repurposing its mobile-first AI technology for data centers, Qualcomm is taking a unique approach to address the growing concern over AI inference costs.
The global investment in AI infrastructure is surging, with projections suggesting nearly $6.7 trillion in capital spending on data centers through 2030. Qualcomm’s focus on performance per watt and compatibility with major AI frameworks positions it to attract customers looking for flexible and cost-effective solutions.
Saudi Arabian AI company Humain has been named the first customer for the new chips. The company plans to deploy 200 megawatts of computing power based on Qualcomm’s technology, with deployments starting in 2026. This partnership underscores the market’s readiness for new players and innovative solutions in the rapidly expanding AI sector.